Xu 2011

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Information Sciences 181 (2011) 2128–2138

Contents lists available at ScienceDirect

Information Sciences
journal homepage: www.elsevier.com/locate/ins

Distance and similarity measures for hesitant fuzzy sets


Zeshui Xu ⇑, Meimei Xia
School of Economics and Management, Southeast University, Nanjing, Jiangsu 211189, China

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we propose a variety of distance measures for hesitant fuzzy sets, based on
Received 29 May 2010 which the corresponding similarity measures can be obtained. We investigate the connec-
Received in revised form 8 January 2011 tions of the aforementioned distance measures and further develop a number of hesitant
Accepted 16 January 2011
ordered weighted distance measures and hesitant ordered weighted similarity measures.
Available online 1 February 2011
They can alleviate the influence of unduly large (or small) deviations on the aggregation
results by assigning them low (or high) weights. Several numerical examples are provided
Keywords:
to illustrate these distance and similarity measures.
Hesitant fuzzy set
Distance measure
Ó 2011 Elsevier Inc. All rights reserved.
Similarity measure
Decision making

1. Introduction

When people make a decision, they are usually hesitant and irresolute for one thing or another which makes it difficult to
reach a final agreement. For example, two decision makers discuss the membership degree of an element x to a set A, and one
wants to assign 0.4 but the other 0.7. Accordingly, the difficulty of establishing a common membership degree is not because
we have a margin of error (intuitionistic fuzzy set (IFS) [1]), or some possibility distribution values (type-2 fuzzy set (T-2FS)
[8,27]), but because we have a set of possible values. To deal with such cases, Torra [31] and Torra and Narukawa [32] intro-
duced the concept of hesitant fuzzy set (HFS) considered as a generalization of fuzzy set (FS) [53]. He discussed the relation-
ships among HFSs and other generalizations of FSs such as IFSs, T-2FSs and fuzzy multiset (FM) [26,47]. He noted that (1) the
envelope of a HFS can be considered as an IFS characterized by a membership degree and a non-membership degree; (2) all
HFSs are T-2FSs in which the membership degree of a given element is defined as a FS; and (3) HFSs and FMs are of the same
form but have different operations.
Distance and similarity measures are fundamentally important in a variety of scientific fields such as decision making,
pattern recognition, machine learning and market prediction, lots of studies have been done on this issue [3,5,18,33]. Among
them, the most widely used distance measures for two FSs A and B on X = {x1, x2, . . ., xn} are the following [6,7,13]:
Pn
 the Hamming distance: dh ðA; BÞ ¼ i¼1 jlA ðxi Þ  lB ðxi Þj;
P
 the normalized Hamming distance: dnh ðA; BÞ ¼ 1n ni¼1 jlA ðxi Þ  lB ðxi Þj;
P 1=2
n 2
 the Euclidean distance: de ðA; BÞ ¼ i¼1 jlA ðxi Þ lB ðxi Þj ;
1=2
Pn
 the normalized Euclidean distance: dne ðA; BÞ ¼ 1n i¼1 j l A ðxi Þ  lB ðxi Þj2 ;
 the Hausdorff metric: dhd(A, B) = maxjlA(xi)  lB(xi)j,

⇑ Corresponding author. Tel.: +86 25 84483382.


E-mail addresses: xu_zeshui@263.net (Z. Xu), meimxia@163.com (M. Xia).

0020-0255/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.ins.2011.01.028
Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138 2129

where lA(xi) and lB(xi) are the membership functions of A and B, respectively, with the condition that
0 6 lA(xi), lB(xi) 6 1, for xi 2 X, i = 1, 2, . . ., n.

Later, a number of other extensions of the above distance measures have been developed for linguistic fuzzy sets (LFSs)
[38] and IFSs [10–12,15–17,24,29,30,35,39–44]. For example, based on the Hamming distance, Xu [38] introduced the con-
cepts of deviation degrees and similarity degrees between two linguistic values, and between two linguistic preference rela-
tions, respectively. Li and Cheng [15] generalized the Hamming distance and the Euclidean distance by adding a parameter
and gave a similarity formula for IFSs only based on the membership degrees and non-membership degrees. Some authors
[17,24,29,30,35,39] improved Li and Cheng’ method [15]. Hung and Yang [11] and Grzegorzewski [10] suggested a lot of sim-
ilarity measures for IFSs and interval-valued fuzzy sets based on the Hausdorff metric. Xu and Chen [44] gave a comprehen-
sive overview of distance and similarity measures for IFSs and developed several continuous distance and similarity
measures for IFSs. Wu and Mendel [37] generalized Jaccard’s similarity measure for T-2FSs [20] and proposed a new simi-
larity measure for interval T-2FSs and compared it with the existing five methods [4,9,25,36,54]. Yang and Lin [51] gave a few
similarity and inclusion measures for T-2FSs and combined them with Yang and Shih’ algorithms [52] as a clustering method
for type-2 fuzzy data.
The aforementioned measures, however, cannot be used to deal with the distance and similarity between HFSs. Due to the
fact that hesitancy is a very common problem in decision making as mentioned earlier, it is necessary to develop some mea-
sures for HFSs. To do this, the remainder of the paper is organized as follows. Section 2 presents the axioms for distance and
similarity measures, gives a variety of distance measures for HFSs and applies them to multi-attribute decision-making with
the known weight information on attributes. In Section 3, we propose a class of ordered weighted distance and similarity
measures under hesitant fuzzy environment, and give several methods to determine the weighting vectors associated with
these distance and similarity measures. Section 4 concludes the paper with some remarks.

2. Distance and similarity measures for HFSs

Hesitant fuzzy sets (HFSs) were first introduced by Torra [31] and Torra and Narukawa [32], it permits the member-
ship degree of an element to a set to be represented as several possible values between 0 and 1. HFSs are very useful in
dealing with the situations where people have hesitancy in providing their preferences over objects in a decision-making
process.

Definition 1 ([31,32]). Let X be a fixed set, a HFS on X is in terms of a function that when applied to X returns a subset of
[0, 1], which can be represented as the following mathematical symbol:

E ¼ fhx; hE ðxÞijx 2 Xg ð1Þ

where hE(x) is a set of values in [0, 1], denoting the possible membership degrees of the element x 2 X to the set E. For con-
venience, we call hE(x) a hesitant fuzzy element (HFE).
It is noted that the number of values in different HFEs may be different, let l(hE(x)) be the number of values in hE(x). We
rðjÞ
arrange the elements in hE(x) in decreasing order, and let hE ðxÞ be the jth largest value in hE(x).
Distance and similarity measures have attracted a lot of attention in the last few decades due to the fact that they can be
applied to many areas such as pattern recognition [15], cluster analysis [48], approximate reasoning [34], image processing
[28], medical diagnosis [30] and decision making [42]. A lot of distance and similarity measures have been developed for FSs,
IFSs, LFSs, T-2FSs, and FMs as mentioned in the introduction, but there is little research on HFSs. Consequently, it is very
necessary to develop some distance and similarity measures under hesitant fuzzy environment. We first address this issue
by putting forward the axioms for distance and similarity measures.

Definition 2. Let M and N be two HFSs on X = {x1, x2, . . ., xn}, then the distance measure between M and N is defined as d(M, N),
which satisfies the following properties:

(1) 0 6 d(M, N) 6 1;
(2) d(M, N) = 0 if and only if M = N;
(3) d(M, N) = d(N, M).

Definition 3. Let M and N be two HFSs on X = {x1, x2, . . ., xn}, then the similarity measure between M and N is defined as
s(M, N), which satisfies the following properties:

(1) 0 6 s(M, N) 6 1;
(2) s(M, N) = 1 if and only if M = N;
(3) s(M, N) = s(N, M).
2130 Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138

By analyzing Definitions 2 and 3, it is noted that s(M, N) = 1  d(M, N), accordingly, we mainly discuss the distance mea-
sures for HFSs in this paper, and the corresponding similarity measures can be obtained easily.
In most cases, l(hM(xi)) – l(hN(xi)), and for convenience, let lxi ¼ maxflðhM ðxi ÞÞ; lðhN ðxi ÞÞg for each xi in X. To operate cor-
rectly, we should extend the shorter one until both of them have the same length when we compare them. To extend the
shorter one, the best way is to add the same value several times in it. In fact, we can extend the shorter one by adding
any value in it. The selection of this value mainly depends on the decision makers’ risk preferences. Optimists anticipate
desirable outcomes and may add the maximum value, while pessimists expect unfavorable outcomes and may add the min-
imum value. For example, let hM(xi) = {0.1, 0.2, 0.3}, hN (xi) = {0.4, 0.5}, and l(hM(xi)) > l(hN(xi)). To operate correctly, we should
extend hN(xi) to hN(xi) = {0.4, 0.4, 0.5} until it has the same length of hM(xi), the optimist may extend hN(xi) as
hN(xi) = {0.4, 0.5, 0.5} and the pessimist may extend it as hN(xi) = {0.4, 0.4, 0.5}. Although the results may be different if we ex-
tend the shorter one by adding different values, this is reasonable because the decision makers’ risk preferences can directly
influence the final decision. The same situation can also be found in many existing Refs. [19,21,22]. In this paper, we assume
that the decision makers are all pessimistic (other situations can be studied similarly).
Drawing on the well-known Hamming distance and the Euclidean distance, we define a hesitant normalized Hamming
distance:
2 3
l xi  
1X n
4 1 X  rðjÞ rðjÞ 
dhnh ðM; NÞ ¼ h ðxi Þ  hN ðxi Þ5 ð2Þ
n i¼1 lxi j¼1 M

and a hesitant normalized Euclidean distance:


2 0 131=2
l xi  2
1 Xn
1 X  rðjÞ rðjÞ 
dhne ðM; NÞ ¼ 4 @ h ðxi Þ  hN ðxi Þ A5 ð3Þ
n i¼1 lxi j¼1 M
rðjÞ rðjÞ
where hM ðxi Þ and hN ðxi Þ are the jth largest values in hM(xi) and hN(xi), respectively, which will be used thereafter.
We can further extend Eqs. (2) and (3) into a generalized hesitant normalized distance:
2 0 131=k
l xi  k
1 X
n X
dghn ðM; NÞ ¼ 4 @1  rðjÞ rðjÞ  A5
hM ðxi Þ  hN ðxi Þ ð4Þ
n i¼1
lxi j¼1

where k > 0.
Eq. (4) is motivated by the generalized idea provided by Yager [49], which has been widely applied in decision making
[2,21,46,55,56].
Especially, if k = 1, then the generalized hesitant normal distance is reduced to the hesitant normalized Hamming dis-
tance; If k = 2, then it is reduced to the hesitant normalized Euclidean distance.
If we apply the Hausdorff metric to the distance measure, then a generalized hesitant normalized Hausdorff distance is
given as
" #
1X n  k 1=k
 rðjÞ rðjÞ 
dghnh ðM; NÞ ¼ max hM ðxi Þ  hN ðxi Þ ð5Þ
n i¼1 j
where k > 0.
Now we discuss two special cases of the generalized hesitant normalized Hausdorff distance:

(1) If k = 1, then Eq. (5) becomes a hesitant normalized Hamming–Hausdorff distance:


1X n  
 rðjÞ rðjÞ 
dhnhh ðM; NÞ ¼ max hM ðxi Þ  hN ðxi Þ ð6Þ
n i¼1 j
(2) If k = 2, then Eq. (5) becomes a hesitant normalized Euclidean–Hausdorff distance:
" #
1X n  2 1=2
 rðjÞ rðjÞ 
dhneh ðM; NÞ ¼ max hM ðxi Þ  hN ðxi Þ ð7Þ
n i¼1 j

Combining the above equations, we define a hybrid hesitant normalized Hamming distance, a hybrid hesitant normalized
Euclidean distance, and a generalized hybrid hesitant normalized distance as follows, respectively:
0 1
lxi    
1 X n
@ 1 X  rðjÞ r ðjÞ   r ðjÞ rðjÞ 
dhhne ðM; NÞ ¼ h ðxi Þ  hN ðxi Þ þ max hM ðxi Þ  hN ðxi ÞA ð8Þ
2n i¼1 lxi j¼1 M j

2 0 131=2
l xi  2  2
1 Xn
1 X  r ðjÞ r ðjÞ   rðjÞ rðjÞ 
dghhn ðM; NÞ ¼ 4 @ h ðxi Þ  hN ðxi Þ þ max hM ðxi Þ  hN ðxi Þ A5 ð9Þ
2n i¼1 lxi j¼1 M j
Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138 2131

2 0 13k
l xi  k  k
1 Xn
1 X  r ðjÞ r ðjÞ   r ðjÞ r ðjÞ 
dhhnh ðM; NÞ ¼ 4 @ h ðxi Þ  hN ðxi Þ þ max hM ðxi Þ  hN ðxi Þ A5 ð10Þ
2n i¼1 lxi j¼1 M j

where k > 0.
Usually, the weight of each element xi 2 X should be taken into account, and so, we present the following weighted dis-
tance measures for HFSs.
P
Assume that the weight of the element xi 2 X is wi (i = 1, 2, . . ., n) with wi 2 [0, 1] and ni¼1 wi ¼ 1, then we get a generalized
hesitant weighted distance:
2 0 131=k
l xi  k
X
n
1 X  r ðjÞ rðjÞ 
dghw ðM; NÞ ¼ 4 wi @ h ðxi Þ  hN ðxi Þ A5 ð11Þ
i¼1
lxi j¼1 M

and a generalized hesitant weighted Hausdorff distance:


" #1=k
Xn  k
 rðjÞ rðjÞ 
dghwh ðM; NÞ ¼ wi max hM ðxi Þ  hN ðxi Þ ð12Þ
j
i¼1

where k > 0.
In particular, if k = 1, then we obtain a hesitant weighted Hamming distance:
2 3
l xi  
X
n
1 X  rðjÞ r ðjÞ 
dhwh ðM; NÞ ¼ wi 4 h ðxi Þ  hN ðxi Þ5 ð13Þ
i¼1
lxi j¼1 M

and a hesitant weighted Hamming–Hausdorff distance:


X
n  
 rðjÞ rðjÞ 
dhwhh ðM; NÞ ¼ wi max hM ðxi Þ  hN ðxi Þ ð14Þ
j
i¼1

If k = 2, then we get a hesitant weighted Euclidean distance:


2 0 131=2
l xi  2
X
n
1 X  r ðjÞ rðjÞ 
dhwe ðM; NÞ ¼ 4 wi @ h ðxi Þ  hN ðxi Þ A5 ð15Þ
i¼1
lxi j¼1 M

and a hesitant weighted Euclidean–Haudorff distance:


" #1=2
Xn  2
 rðjÞ rðjÞ 
dhweh ðM; NÞ ¼ wi max hM ðxi Þ  hN ðxi Þ ð16Þ
j
i¼1

Furthermore, we develop a generalized hybrid hesitant weighted distance combining the generalized hesitant weighted
distance and the generalized hesitant weighted Hausdorff distance as
2 0 13k
lxi  k  k
X n
1 X  rðjÞ rðjÞ   rðjÞ rðjÞ 
dghhw ðM; NÞ ¼ 4 wi @ h ðxi Þ  hN ðxi Þ þ max hM ðxi Þ  hN ðxi Þ A5 ð17Þ
i¼1
lxi j¼1 M j

where k > 0.
In the special cases where k = 1, 2, Eq. (17) is reduced to a hybrid hesitant weighted Hamming distance and a hybrid hes-
itant weighted Euclidean distance as follows, respectively:
0 1
l xi    
X
n
1 X  r ðjÞ r ðjÞ   rðjÞ rðjÞ 
dhhwh ðM; NÞ ¼ wi @ h ðxi Þ  hN ðxi Þ þ max hM ðxi Þ  hN ðxi ÞA ð18Þ
i¼1
lxi j¼1 M j

2 0 131=2
lxi  2  2
X n
1 X  r ðjÞ r ðjÞ   rðjÞ rðjÞ 
dhhwe ðM; NÞ ¼ 4 wi @ h ðxi Þ  hN ðxi Þ þ max hM ðxi Þ  hN ðxi Þ A5 ð19Þ
i¼1
lxi j¼1 M j

We find that all the above distance measures are discrete, if both the universe of discourse and the weight of element are
Rb
continuous, and the weight of x 2 X = [a, b] is w(x), where w(x) 2 [0, 1] and a wðxÞdx ¼ 1, then we define a continuous hes-
itant weighted Hamming distance, a continuous hesitant weighted Euclidean distance and a generalized continuous hesitant
weighted distance as follows, respectively:
2132 Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138

Z !
lx  
b
1X  rðjÞ rðjÞ 
dchwh ðM; NÞ ¼ wðxÞ h ðxÞ  hN ðxÞ dx ð20Þ
a lx j¼1 M
"Z ! #1=2
lx  2
b
1X  rðjÞ rðjÞ 
dchwe ðM; NÞ ¼ wðxÞ hM ðxÞ  hN ðxÞ dx ð21Þ
a lx j¼1
"Z ! #1=k
lx  k
b
1X  rðjÞ rðjÞ 
dgchw ðM; NÞ ¼ wðxÞ hM ðxÞ  hN ðxÞ dx ð22Þ
a lx j¼1

where k > 0.
If w(x) = 1/(b  a), for any x 2 [a, b], then the continuous hesitant weighted Hamming distance is reduced to a continuous
hesitant normalized Hamming distance:
Z !
lx  
1 b
1X  rðjÞ rðjÞ 
dchnh ðM; NÞ ¼ h ðxÞ  hN ðxÞ dx ð23Þ
ba a lx j¼1 M

while the continuous hesitant weighted Euclidean distance is reduced to a continuous hesitant normalized Euclidean
distance:
" Z ! #1=2
lx  2
1 b
1X  rðjÞ rðjÞ 
dchne ðM; NÞ ¼ h ðxÞ  hN ðxÞ dx ð24Þ
ba a lx j¼1 M

and the generalized continuous hesitant weighted distance is reduced to a generalized continuous hesitant normalized
distance:
" Z ! #1=k
lx  k
1 b
1X  rðjÞ rðjÞ 
dgchw ðM; NÞ ¼ hM ðxÞ  hN ðxÞ dx ð25Þ
ba a lx j¼1

where k > 0.
Using the traditional Hausdorff metric, we define a generalized continuous hesitant weighted Hausdorff distance as
"Z #1=k
b  k
 rðjÞ rðjÞ 
dgchwh ðM; NÞ ¼ wðxÞ max hM ðxÞ  hN ðxÞ dx ð26Þ
a j

where k > 0.
In the special cases where k = 1, 2, the generalized continuous hesitant weighted distance is reduced to a continuous hes-
itant weighted Hamming–Hausdorff distance:
Z b  
 rðjÞ rðjÞ 
dchwhh ðM; NÞ ¼ wðxÞ max hM ðxÞ  hN ðxÞdx ð27Þ
a j

and a continuous hesitant weighted Euclidean–Hausdorff distance:


"Z #1=2
b  2
 rðjÞ rðjÞ 
dchweh ðM; NÞ ¼ wðxÞ max hM ðxÞ  hN ðxÞ dx ð28Þ
a j

respectively.
If w(x) = 1/(b  a), for any x 2 [a, b], then the generalized continuous hesitant weighted distance becomes a generalized
continuous hesitant normalized distance:
" Z #1=k
1 b  k
 rðjÞ rðjÞ 
dgchn ðM; NÞ ¼ max hM ðxÞ  hN ðxÞ dx ð29Þ
ba a j

where k > 0, while the continuous hesitant weighted Euclidean–Hausdorff distance becomes a continuous hesitant normal-
ized Hamming–Hausdorff distance:
Z b  
1  rðjÞ rðjÞ 
dchneh ðM; NÞ ¼ max hM ðxÞ  hN ðxÞdx ð30Þ
ba a j

and the continuous hesitant weighted Euclidean–Hausdorff distance becomes a continuous hesitant normalized Euclidean–
Hausdorff distance:
Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138 2133

" Z #1=2
1 b  2
 rðjÞ rðjÞ 
dchneh ðM; NÞ ¼ max hM ðxÞ  hN ðxÞ dx ð31Þ
ba a j

Analogous to the generalized hybrid hesitant weighted distance, we develop a generalized hybrid continuous hesitant
weighted distance as
"Z li 
!#1=k
b
1 X k 1  k
 rðjÞ rðjÞ   rðjÞ rðjÞ 
dghchw ðM; NÞ ¼ wðxÞ hM ðxÞ  hN ðxÞ þ max hM ðxÞ  hN ðxÞ ð32Þ
a 2li j¼1 2 j

where k > 0.
If w(x) = 1/(b  a), for any x 2 [a, b], then the generalized hybrid continuous hesitant weighted distance becomes a gener-
alized hybrid continuous hesitant normalized distance:
" Z !#1=k
lx  k 1  k
1 b
1 X  rðjÞ rðjÞ   rðjÞ rðjÞ 
dghchn ðM; NÞ ¼ h ðxÞ  hN ðxÞ þ max hM ðxÞ  hN ðxÞ ð33Þ
ba a 2lx j¼1 M 2 j

where k > 0.
Let k = 1, 2, then from Eq. (32), we get a hybrid continuous hesitant weighted Hamming distance and a continuous hybrid
continuous hesitant weighted Euclidean distance as
Z !
lx   1  
b
1 X  rðjÞ rðjÞ   rðjÞ rðjÞ 
dhchwh ðM; NÞ ¼ wðxÞ hM ðxÞ  hN ðxÞ þ max hM ðxÞ  hN ðxÞ ð34Þ
a 2lx j¼1 2 j

and
"Z !#1=2
lx  2 1  2
b
1 X  rðjÞ rðjÞ   rðjÞ rðjÞ 
dhchwe ðM; NÞ ¼ wðxÞ hM ðxÞ  hN ðxÞ þ max hM ðxÞ  hN ðxÞ ð35Þ
a 2lx j¼1 2 j

respectively.
Let w(x) = 1/(b  a), for any x 2 [a, b], then Eqs. (34) and (35) are reduced to a hybrid continuous hesitant normalized
Hamming distance:
Z !
lx   1  
1 b
1 X  rðjÞ rðjÞ   rðjÞ rðjÞ 
dhchnh ðM; NÞ ¼  h ðxÞ  h ðxÞ  þ max h
 M ðxÞ  h ðxÞ  ð36Þ
ba a 2lx j¼1 M N
2 j N

and a hybrid continuous hesitant normalized Euclidean distance:


" Z !#1=2
lx  2 1  2
1 b
1 X  rðjÞ rðjÞ   rðjÞ rðjÞ 
dhchne ðM; NÞ ¼ hM ðxÞ  hN ðxÞ þ max hM ðxÞ  hN ðxÞ ð37Þ
ba a 2lx j¼1 2 j

respectively.
From the foregoing analysis, we find that the generalized hesitant weighted distance, the generalized hesitant weighted
Hausdorff distance and the generalized hybrid hesitant weighted distance are three fundamental distance measures, based
on which all of the other developed distance measures can be obtained under some special conditions.
In what follows, we give an example adapted from [14] to illustrate our distance measures for HFSs:

Example 1. Energy is an indispensable factor for the socio-economic development of societies. Thus the correct energy
policy affects economic development and environment, and so, the most appropriate energy policy selection is very
important. Suppose that there are five alternatives (energy projects) Ai (i = 1, 2, 3, 4, 5) to be invested, and four attributes to be
considered: P1: technological; P2: environmental; P3: socio-political; P4: economic (more details about them can be found in
[14]). The attribute weight vector is w = (0.15, 0.3, 0.2, 0.35)T. Several decision makers are invited to evaluate the performance
of the five alternatives. For an alternative under an attribute, although all of the decision makers provide their evaluated
values, some of these values may be repeated. However, a value repeated more times does not indicate that it has more
importance than other values repeated less times. For example, the value repeated one time may be provided by a decision
maker who is an expert at this area, and the value repeated twice may be provided by two decision makers who are not
familiar with this area. In such cases, the value repeated one time may be more important than the one repeated twice. To
get a more reasonable result, it is better that the decision makers give their evaluations anonymously. We only collect all of
the possible values for an alternative under an attribute, and each value provided only means that it is a possible value, but
its importance is unknown. Thus the times that the values repeated are unimportant, and it is reasonable to allow these
values repeated many times appear only once. The HFS is just a tool to deal with such cases, and all possible evaluations for
an alternative under the attributes can be considered as a HFS. The results evaluated by the decision makers are contained in
a hesitant fuzzy decision matrix, shown in Table 1.
2134 Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138

Table 1
Hesitant fuzzy decision matrix.

P1 P2 P3 P4
A1 {0.5, 0.4, 0.3} {0.9, 0.8, 0.7, 0.1} {0.5, 0.4, 0.2} {0.9, 0.6, 0.5, 0.3}
A2 {0.5, 0.3} {0.9, 0.7, 0.6, 0.5, 0.2} {0.8, 0.6, 0.5, 0.1} {0.7, 0.3, 0.4}
A3 {0.7, 0.6} {0.9, 0.6} {0.7, 0.5, 0.3} {0.6, 0.4}
A4 {0.8, 0.7, 0.4, 0.3} {0.7, 0.4, 0.2} {0.8, 0.1} {0.9, 0.8, 0.6}
A5 {0.9, 0.7, 0.6, 0.3, 0.1} {0.8, 0.7, 0.6, 0.4} {0.9, 0.8, 0.7} {0.9, 0.7, 0.6, 0.3}

Suppose that the ideal alternative is A⁄ = {1} seen as a special HFS, we can calculate the distance between each alternative
and the ideal alternative using our distance measures.
If we use the generalized hesitant weighted distance, the generalized hesitant Hausdorff distance, and the generalized
hybrid hesitant weighted distance to calculate the deviations between each alternative and the ideal alternative, then we get
the rankings of these alternatives, which are listed in Tables 2–4, respectively, when some values of the parameter are given.
We find that the rankings are different as the parameter k (which can be considered as the decision makers’ risk attitude)
changes, consequently, the proposed distance measures can provide the decision makers more choices as the different values
of the parameter are given according to the decision makers’ risk attitudes.

3. Ordered distance and similarity measures for HFSs

Xu and Chen [45] defined several ordered weighted distance measures whose prominent characteristic is that they can
alleviate (or intensify) the influence of unduly large (or small) deviations on the aggregation results by assigning them low
(or high) weights. This desirable characteristic makes the ordered weighted distance measures very useful in many actual
fields such as group decision making, medical diagnosis, data mining, and pattern recognition. Yager [50] generalized Xu
and Chen’ distance measures and provided a variety of ordered weighted averaging norms, based on which he proposed sev-
eral similarity measures. Merigó and Gil-Lafuente [23] introduced an ordered weighted averaging distance operator and gave
its application in the selection of financial products. In what follows, we develop some ordered distance measures for HFSs.
Motivated by the ordered weighted idea [45], we define a hesitant ordered weighted Hamming distance:
2 lxr^ ðiÞ
3
X
n
1 X  
 rðjÞ rðjÞ 
dhowh ðM; NÞ ¼ xi 4 h ðxr^ ðiÞ Þ  hN ðxr^ ðiÞ Þ5 ð38Þ
i¼1
lxr^ ðiÞ j¼1 M

Table 2
Results obtained by the generalized hesitant weighted distance.

A1 A2 A3 A4 A5 Rankings
k=1 0.4799 0.5027 0.4025 0.4292 0.3558 A5  A3  A4  A1  A2
k=2 0.5378 0.5451 0.4366 0.5052 0.4129 A5  A3  A4  A1  A2
k=6 0.6599 0.6476 0.5156 0.6704 0.5699 A3  A5  A2  A1  A4
k = 10 0.7213 0.7046 0.5607 0.7373 0.6537 A3  A5  A2  A1  A4

Table 3
Results obtained by the generalized hesitant weighted Hausdorff distance.

A1 A2 A3 A4 A5 Rankings
k=1 0.7800 0.7700 0.5300 0.6650 0.6200 A3  A5  A4  A2  A1
k=2 0.7849 0.7740 0.5441 0.6953 0.6473 A3  A5  A4  A2  A1
k=6 0.8043 0.7904 0.5889 0.7673 0.7163 A3  A5  A4  A2  A1
k = 10 0.8216 0.8063 0.6156 0.7991 0.7597 A3  A5  A4  A2  A1

Table 4
Results obtained by the generalized hybrid hesitant weighted distance.

A1 A2 A3 A4 A5 Rankings
k=1 0.6300 0.6363 0.4662 0.5471 0.4879 A3  A5  A4  A1  A2
k=2 0.6613 0.6595 0.4903 0.6002 0.5301 A3  A5  A4  A2  A1
k=6 0.7321 0.7190 0.5523 0.7188 0.6431 A3  A5  A4  A2  A1
k = 10 0.7628 0.7475 0.5748 0.7523 0.6850 A3  A5  A2  A4  A1
Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138 2135

and a hesitant ordered weighted Euclidean distance:


2 0 lxr^ ðiÞ
131=2
4
X n
@ 1 X  rðjÞ rðjÞ
2
 A5
dhowe ðM; NÞ ¼ xi h ðxr^ ðiÞ Þ  hN ðxr^ ðiÞ Þ ð39Þ
i¼1
lxr^ ðiÞ j¼1 M

respectively, where r(j) is given as in Section 2, and r


^ : ð1; 2; . . . ; nÞ ! ð1; 2; . . . ; nÞ is a permutation satisfying
lxr^ ðiþ1Þ lxr^ ðiÞ
1 X  rðjÞ rðjÞ

 1 X 
 rðjÞ rðjÞ


hM ðxr^ ðiþ1Þ Þ  hN ðxr^ ðiþ1Þ Þ P hM ðxr^ ðiÞ Þ  hN ðxr^ ðiÞ Þ; i ¼ 1; 2; . . . ; n  1 ð40Þ
lxr^ ðiþ1Þ j¼1
l xr^ ðiÞ j¼1

Generalizing Eqs. (38) and (39), we define a generalized hesitant ordered weighted distance measure:
2 0 lxr^ ðiÞ
131=k
X n
1 X  k
 rðjÞ rðjÞ 
dghow ðM; NÞ ¼ 4 xi @ h ðxr^ ðiÞ Þ  hN ðxr^ ðiÞ Þ A5 ð41Þ
i¼1
lxr^ ðiÞ j¼1 M

where k > 0.
With the Hausdorff metric, we develop a generalized hesitant ordered weighted Hausdorff distance as
" #1=k
Xn  k
dghowh ðM; NÞ ¼ xi max hrMðjÞ ðxr_ ðiÞ Þ  hrNðjÞ ðxr_ ðiÞ Þ ð42Þ
j
i¼1

where k > 0 and r_ : ð1; 2; . . . ; nÞ ! ð1; 2; . . . ; nÞ is a permutation satisfying


   
 rðjÞ rðjÞ   rðjÞ rðjÞ 
max hM ðxr_ ðiþ1Þ Þ  hN ðxr_ ðiþ1Þ Þ P max hM ðxr_ ðiÞ Þ  hN ðxr_ ðiÞ Þ; i ¼ 1; 2; . . . ; n  1 ð43Þ
j j

In what follows, we discuss two special cases of the generalized hesitant ordered weighted Hausdorff distance:

(1) If k = 1, then Eq. (42) is reduced to a hesitant ordered weighted Hamming–Hausdorff distance:

X
n  
dghowhh ðM; NÞ ¼ xi max hrMðjÞ ðxr_ ðiÞ Þ  hrNðjÞ ðxr_ ðiÞ Þ ð44Þ
j
i¼1

(2) If k = 2, then Eq. (42) is reduced to a hesitant ordered weighted Euclidean–Hausdorff distance:
" #1=2
X
n  2
dghoweh ðM; NÞ ¼ xi max hrMðjÞ ðxr_ ðiÞ Þ  hrNðjÞ ðxr_ ðiÞ Þ ð45Þ
j
i¼1

Combining Eqs. (41) and (42), we develop a generalized hybrid hesitant ordered weighted distance as
2 0 lxr€ ðiÞ
131=k
X n
1 X  k 1  k
 r ðjÞ r ðjÞ   rðjÞ rðjÞ 
dghhow ðM; NÞ ¼ 4 xi @ h ðxr€ ðiÞ Þ  hN ðxr€ ðiÞ Þ þ max hM ðxr€ ðiÞ Þ  hN ðxr€ ðiÞ Þ A5 ð46Þ
i¼1
2lxr€ ðiÞ j¼1 M 2 j

where k > 0; r
€ : ð1; 2; . . . ; nÞ ! ð1; 2; . . . ; nÞ is a permutation such that

lxr€ ðiþ1Þ
1 X  rðjÞ rðjÞ
 1


 rðjÞ rðjÞ


hM ðxr€ ðiþ1Þ Þ  hN ðxr€ ðiþ1Þ Þ þ max hM ðxr€ ðiþ1Þ Þ  hN ðxr€ ðiþ1Þ Þ
2lxr€ ðiþ1Þ j¼1
2 j

lxr€ ðiÞ
1 X  rðjÞ rðjÞ
 1


 rðjÞ rðjÞ


P hM ðxr€ ðiÞ Þ  hN ðxr€ ðiÞ Þ þ max hM ðxr€ ðiÞ Þ  hN ðxr€ ðiÞ Þ; i ¼ 1; 2; . . . ; n  1 ð47Þ
2lxr€ ðiÞ j¼1 2 j

As the parameter and the weight vector change, some special cases can be obtained just as discussed in Sections 2 and 3.
Let do denote the ordered distance measures defined above, then the ordered similarity measures for HFSs can be given as
so = 1  do.
Another important issue is the determination of the weight vectors associated with the ordered weighted distance mea-
sures. Inspired by Xu and Chen [45], below we give three ways to determine the weight vectors.
Considering each element in M and N as a special HFS, d(hM(xq(i)), hN(xq(i))) (i = 1, 2, . . ., n) as given in Section 2, and denot-
ing r^ ; r_ and r
€ as q, we have
2136 Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138

Table 5
Results obtained by the generalized hesitant ordered weighted distance.

A1 A2 A3 A4 A5 Rankings
k=1 0.5085 0.5132 0.4014 0.4565 0.3600 A5  A3  A4  A1  A2
k=2 0.5584 0.5545 0.4327 0.5304 0.4222 A5  A3  A4  A2  A1
k=6 0.6604 0.6561 0.5149 0.6863 0.5915 A3  A5  A2  A1  A4
k = 10 0.7160 0.7140 0.5639 0.7492 0.6771 A3  A5  A2  A1  A4

(1) Let

dðhM ðxqðiÞ Þ; hN ðxqðiÞ ÞÞ


xi ¼ Pn ; i ¼ 1; 2; . . . ; n ð48Þ
k¼1 dðhM ðxqðkÞ Þ; hN ðxqðkÞ ÞÞ
Pn
then xi+1 P xi P 0, i = 1, 2, . . ., n  1, and i¼1 xi ¼ 1.
(2) Let

edðhM ðxqðiÞ Þ;hN ðxqðiÞ ÞÞ


xi ¼ Pn dðhM ðxqðkÞ Þ;hN ðxqðkÞ ÞÞ
; i ¼ 1; 2; . . . ; n ð49Þ
k¼1 e
Pn
then 0 6 xi+1 6 xi, i = 1, 2, . . ., n  1, and i¼1 xi ¼ 1.
(3) Let

X
n
_ M ; hN Þ ¼ 1
dðh dðhM ðxqðkÞ Þ; hN ðxqðkÞ ÞÞ ð50Þ
n k¼1

and
 
 1X n 
€ _  
dðdðh ðx
M qðiÞ Þ; h ðx
N qðiÞ ÞÞ; dðh M ; h N ÞÞ ¼ dðh ðx
M qðiÞ Þ; h ðx
N qðiÞ ÞÞ  dðh ðx
M qðkÞ Þ; h ðx ÞÞ
N qðkÞ  ð51Þ
 n k¼1 

then we define

€ dðh
1  dð _ M ; hN Þ; dðhM ðx Þ; hN ðx ÞÞÞ
qðiÞ qðiÞ
xi ¼ Pn € _
k¼1 ð1  dðdðhM ; hN Þ; dðhM ðxqðiÞ Þ; hN ðxqðiÞ ÞÞÞÞ
 P 
1  dðhM ðxqðiÞ Þ; hN ðxqðiÞ ÞÞ  1n nk¼1 dðhM ðxqðkÞ Þ; hN ðxqðkÞ ÞÞ
¼ Pn   Pn  ; i ¼ 1; 2; . . . ; n ð52Þ
 1 
k¼1 1  dðhM ðxqðiÞ Þ; hN ðxqðiÞ ÞÞ  n k¼1 dðhM ðxqðkÞ Þ; hN ðxqðkÞ ÞÞ
P
from which we get xi P 0, i = 1, 2, . . ., n, and ni¼1 xi ¼ 1.
We find that the weight vector derived from Eq. (48) is a monotonic decreasing sequence, the weight vector derived from
Eq. (49) is a monotonic increasing sequence, and the weight vector derived from Eq. (52) combine the above two cases, i.e.,
P
the closer the value d(hM(xq(i)), hN(xq(i))) to the mean 1n nk¼1 dðhM ðxqðkÞ Þ; hN ðxqðkÞ ÞÞ, the larger the weight xi.
In Example 1, if the attribute weight vector is unknown, then we can use the ordered weighted distance measures to cal-
culate the distance between each alternative and the ideal alternative. Without loss of generality, suppose that d = dhnh in Eq.
(48) and q is given as in Eq. (40), we use the generalized hesitant ordered weighted distance measure (41) to calculate the
distance between each alternative and the ideal alternative. The derived results are shown in Table 5 with the different val-
ues of the parameter.

4. Concluding remarks

In this paper, we have investigated the distance measures for HFSs. Based on ideas of the well-known Hamming distance,
the Euclidean distance, the Hausdorff metric and their generalizations, we have developed a class of hesitant distance mea-
sures, and discussed their properties and relations as their parameters change. We have also given a variety of ordered
weighted distance measures for HFSs in which the distances are rearranged in decreasing order, and given three ways to
determine the associated weighting vectors. With the relationship between distance measures and similarity measures,
the corresponding similarity measures for HFSs have been obtained. It should be pointed out that all of the above measures
are based on the assumption that if the corresponding HFEs in HFSs do not have the same length, then the shorter one should
be extended by adding the minimum value in it until both the HFEs have the same length. In fact, we can extend the shorter
HFE by adding any value in it until it has the same length of the longer one according to the decision makers’ preferences and
actual situations.
Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138 2137

Acknowledgments

The work was partly supported by the National Science Fund for Distinguished Young Scholars of China (No. 70625005),
the National Natural Science Foundation of China (No. 71071161), the Program Sponsored for Scientific Innovation Research
of College Graduate in Jiangsu Province (No. CX10B_059Z), and the Scientific Research Foundation of Graduate School of
Southeast University. The authors are very grateful to the Editor-in-Chief, Professor Witold Pedrycz, and the anonymous ref-
erees, for their constructive comments and suggestions that led to an improved version of this paper.

References

[1] K. Atanassov, Intuitionistic fuzzy sets, Fuzzy Sets and Systems 20 (1986) 87–96.
[2] G. Beliakov, Learn weights in the generalized OWA operator operators, Fuzzy Optimization and Decision Making 4 (2005) 119–130.
[3] J.J. Buckley, Y. Hayashi, Fuzzy input–output controllers are universal approximates, Fuzzy Sets and Systems 58 (1993) 273–278.
[4] H. Bustince, Indicator of inclusion grade for interval-valued fuzzy sets: application to approximate reasoning based on interval-valued fuzzy sets,
International Journal of Approximate Reasoning 23 (2000) 137–209.
[5] K.S. Candan, W.S. Li, M.L. Priya, Similarity-based ranking and query processing in multimedia databases, Data and Knowledge Engineering 35 (2000)
259–298.
[6] B.B. Chaudhuri, A. Rosenfeld, A modified Hausdorff distance between fuzzy sets, Information Sciences 118 (1999) 159–171.
[7] P. Diamond, P. Kloeden, Metric Spaces of Fuzzy Sets Theory and Applications, World Scientific Publishing, Singapore, 1994.
[8] D. Dubois, H. Prade, Fuzzy Sets and Systems: Theory and Applications, Academic Press, New York, 1980.
[9] M.B. Gorzalczany, A method of inference in approximate reasoning based on interval-valued fuzzy sets, Fuzzy Sets and Systems 21 (1987) 1–17.
[10] P. Grzegorzewski, Distances between intuitionistic fuzzy sets and/or interval-valued fuzzy sets based on the Hausdorff metric, Fuzzy Sets and Systems
148 (2004) 319–328.
[11] W.L. Hung, M.S. Yang, Similarity measures of intuitionistic fuzzy sets based on Hausdorff distance, Pattern Recognition Letters 25 (2004) 1603–1611.
[12] W.L. Hung, M.S. Yang, Similarity measures of intuitionistic fuzzy sets based on Lp metric, International Journal of Approximate Reasoning 46 (2007)
120–136.
[13] J. Kacprzyk, Multistage Fuzzy Control, Wiley, Chichester, 1997.
[14] C. Kahraman, I. Kaya, A fuzzy multicriteria methodology for selection among energy alternatives, Expert Systems with Applications 37 (2010) 6270–
6281.
[15] D.F. Li, C.T. Cheng, New similarity measures of intuitionistic fuzzy sets and application to pattern recognitions, Pattern Recognition Letters 23 (2002)
221–225.
[16] Y.H. Li, D.L. Olson, Z. Qin, Similarity measures between intuitionistic fuzzy (vague) sets: a comparative analysis, Pattern Recognition Letters 28 (2007)
278–285.
[17] Z.Z. Liang, P.F. Shi, Similarity measures on intuitionistic fuzzy sets, Pattern Recognition Letters 24 (2003) 2687–2693.
[18] X.C. Liu, Entropy, distance measure and similarity measure of fuzzy sets and their relations, Fuzzy Sets and Systems 52 (1992) 305–318.
[19] H.W. Liu, G.J. Wang, Multi-criteria decision making methods based on intuitionistic fuzzy sets, European Journal of Operational Research 197 (2007)
220–233.
[20] J.M. Mendel, Computing with words, when words can mean different things to different people, in: Proceedings third International ICSC Symposium on
Fuzzy Logic and Applications, Rochester, NY, 1999, pp. 158–164.
[21] J.M. Merigó, A.M. Gil-Lafuente, The induced generalized OWA operator, Information Sciences 179 (2009) 729–741.
[22] J.M. Merigó, M. Casanovas, Induced aggregation operators in decision making with the Dempster–Shafer belief structure, International Journal of
Intelligent Systems 24 (2009) 934–954.
[23] J.M. Merigó, A.M. Gil-Lafuente, New decision-making techniques and their application in the selection of financial products, Information Sciences 180
(2010) 2085–2094.
[24] H.B. Mitchell, On the Dengfeng–Chuntian similarity measure and its application to pattern recognition, Pattern Recognition Letters 24 (2003) 3101–
3104.
[25] H.B. Mitchell, Pattern recognition using type-II fuzzy sets, Information Sciences 170 (2005) 409–418.
[26] S. Miyamoto, Multisets and fuzzy multisets, in: Z.-Q. Liu, S. Miyamoto (Eds.), Soft Computing and Human-centered Machines, Springer, Berlin, 2000,
pp. 9–33.
[27] S. Miyamoto, Remarks on basics of fuzzy sets and fuzzy multisets, Fuzzy Sets and Systems 156 (2005) 427–431.
[28] S.K. Pal, R.A. King, Image enhancement using smoothing with fuzzy sets, IEEE Transactions on Systems, Man and Cybernetics 11 (1981) 495–501.
[29] E. Szmidt, J. Kacprzyk, Distances between intuitionistic fuzzy sets, Fuzzy Sets and Systems 114 (2000) 505–518.
[30] E. Szmidt, J. Kacprzyk, Intuitionistic fuzzy sets in intelligent data analysis for medical diagnosis, in: V.A. Alexandrov et al. (Eds.), CCS 2001, LNCS, vol.
2074, Springer-Verlag, Berlin, Heidelberg, 2001, pp. 263–271.
[31] V. Torra, Hesitant fuzzy sets, International Journal of Intelligent Systems 25 (2010) 529–539.
[32] V. Torra, Y. Narukawa, On hesitant fuzzy sets and decision, in: The 18th IEEE International Conference on Fuzzy Systems, Jeju Island, Korea, 2009, pp.
1378–1382.
[33] I.B. Turksen, Z. Zhong, An approximate analogical reasoning approach based on similarity measures, IEEE Transactions on Systems, Man and
Cybernetics 18 (1988) 1049–1056.
[34] T.J. Wang, Z.D. Lu, F. Li, Bidirectional approximate reasoning based on weighted similarity measures of vague sets, Journal of Computer Engineering and
Science 24 (2002) 96–100.
[35] W.Q. Wang, X.L. Xin, Distance measure between intuitionistic fuzzy sets, Pattern Recognition Letters 26 (2005) 2063–2069.
[36] D. Wu, J.M. Mendel, A vector similarity measure for linguistic approximation: interval type-2 and type-1 fuzzy sets, Information Sciences 178 (2008)
381–402.
[37] D. Wu, J.M. Mendel, A comparative study of ranking methods, similarity measures and uncertainty measures for interval type-2 fuzzy sets, Information
Sciences 179 (2009) 1169–1192.
[38] Z.S. Xu, Deviation measures of linguistic preference relations in group decision making, Omega 17 (2005) 432–445.
[39] Z.S. Xu, Some similarity measures of intuitionistic fuzzy sets and their applications to multiple attribute decision making, Fuzzy Optimization and
Decision Making 6 (2007) 109–121.
[40] Z.S. Xu, Intuitionistic preference relations and their application in group decision making, Information Sciences 177 (2007) 2363–2379.
[41] Z.S. Xu, J. Chen, J.J. Wu, Clustering algorithm for intuitionistic fuzzy sets, Information Sciences 178 (2008) 3775–3790.
[42] Z.S. Xu, A method based on distance measure for interval-valued intuitionistic fuzzy group decision making, Information Sciences 180 (2010) 181–190.
[43] Z.S. Xu, Choquet integrals of weighted intuitionistic fuzzy information, Information Sciences 180 (2010) 726–736.
[44] Z.S. Xu, J. Chen, An overview of distance and similarity measures of intuitionistic fuzzy sets, International Journal of Uncertainty, Fuzziness and
Knowledge-Based Systems 16 (2008) 529–555.
[45] Z.S. Xu, J. Chen, Ordered weighted distance measure, Journal of Systems Science and Systems Engineering 16 (2008) 529–555.
2138 Z. Xu, M. Xia / Information Sciences 181 (2011) 2128–2138

[46] Z. S Xu, M.M. Xia, Induced generalized intuitionistic fuzzy operators, Knowledge-Based Systems 24 (2011) 197–209.
[47] R.R. Yager, On the theory of bags, International Journal of General Systems 13 (1986) 23–37.
[48] R.R. Yager, On ordered weighted averaging aggregation operators in multi-criteria decision making, IEEE Transactions on Systems, Man, and
Cybernetics18 (1988) 183–190.
[49] R.R. Yager, Generalized OWA aggregation operators, Fuzzy Optimization and Decision Making 3 (2004) 93–107.
[50] R.R. Yager, Norms induced from OWA operators, IEEE Transactions on Fuzzy Systems 18 (2010) 57–66.
[51] M.S. Yang, D.C. Lin, On similarity and inclusion measures between type-2 fuzzy sets with an application to clustering, Computers and Mathematics
with Applications 57 (2009) 896–907.
[52] M.S. Yang, H.M. Shih, Cluster analysis based on fuzzy relations, Fuzzy Sets and Systems 120 (2001) 197–212.
[53] L.A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338–353.
[54] W.Y. Zeng, H.X. Li, Relationship between similarity measure and entropy of interval valued fuzzy sets, Fuzzy Sets and Systems 157 (2006) 1477–1484.
[55] H. Zhao, Z.S. Xu, M.F. Ni, S.S. Liu, Generalized aggregation operators for intuitionistic fuzzy Sets, International Journal of Intelligent Systems 25 (2010)
1–30.
[56] L.G. Zhou, H.Y. Chen, Continuous generalized OWA operator and its application to decision making, Fuzzy Sets and Systems (2010), doi:10.1016/
j.fss.2010.05.009.

You might also like